Filamentary TaOx/HfO2 ReRAM Devices for Neural Networks Training with Analog In‐Memory Computing

The in‐memory computing paradigm aims at overcoming the intrinsic inefficiencies of Von‐Neumann computers by reducing the data‐transport per arithmetic operation. Crossbar arrays of multilevel memristive devices enable efficient calculations of matrix‐vector‐multiplications, an operation extensively...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Advanced electronic materials Jg. 8; H. 10
Hauptverfasser: Stecconi, Tommaso, Guido, Roberto, Berchialla, Luca, La Porta, Antonio, Weiss, Jonas, Popoff, Youri, Halter, Mattia, Sousa, Marilyne, Horst, Folkert, Dávila, Diana, Drechsler, Ute, Dittmann, Regina, Offrein, Bert Jan, Bragaglia, Valeria
Format: Journal Article
Sprache:Englisch
Veröffentlicht: 01.10.2022
Schlagworte:
ISSN:2199-160X, 2199-160X
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Abstract The in‐memory computing paradigm aims at overcoming the intrinsic inefficiencies of Von‐Neumann computers by reducing the data‐transport per arithmetic operation. Crossbar arrays of multilevel memristive devices enable efficient calculations of matrix‐vector‐multiplications, an operation extensively called on in artificial intelligence (AI) tasks. Resistive random‐access memories (ReRAMs) are promising candidate devices for such applications. However, they generally exhibit large stochasticity and device‐to‐device variability. The integration of a sub‐stoichiometric metal‐oxide within the ReRAM stack can improve the resistive switching graduality and stochasticity. To this purpose, a conductive TaOx layer is developed and stacked on HfO2 between TiN electrodes, to create a complementary metal‐oxide‐semiconductor‐compatible ReRAM structure. This device shows accumulative conductance updates in both directions, as required for training neural networks. Moreover, by reducing the TaOx thickness and by increasing its resistivity, the device resistive states increase, as required for reduced power consumption. An electric field‐driven TaOx oxidation/reduction is responsible for the ReRAM switching. To demonstrate the potential of the optimized TaOx/HfO2 devices, the training of a fully‐connected neural network on the Modified National Institute of Standards and Technology database dataset is simulated and benchmarked against a full precision digital implementation. A conductive TaOx material is developed to create a complementary metal‐oxide‐semiconductor‐compatible bilayer TaOx/HfO2 resistive random‐access memory. This device is based on filamentary switching mechanisms, but shows reduced stochasticity and improved graduality compared to metal–insulator–metal baselines. Applying short (<200 ns) and low amplitude (<1.5 V) voltage pulses the device responds with quasi‐analog conductance updates in both directions.
AbstractList The in‐memory computing paradigm aims at overcoming the intrinsic inefficiencies of Von‐Neumann computers by reducing the data‐transport per arithmetic operation. Crossbar arrays of multilevel memristive devices enable efficient calculations of matrix‐vector‐multiplications, an operation extensively called on in artificial intelligence (AI) tasks. Resistive random‐access memories (ReRAMs) are promising candidate devices for such applications. However, they generally exhibit large stochasticity and device‐to‐device variability. The integration of a sub‐stoichiometric metal‐oxide within the ReRAM stack can improve the resistive switching graduality and stochasticity. To this purpose, a conductive TaOx layer is developed and stacked on HfO2 between TiN electrodes, to create a complementary metal‐oxide‐semiconductor‐compatible ReRAM structure. This device shows accumulative conductance updates in both directions, as required for training neural networks. Moreover, by reducing the TaOx thickness and by increasing its resistivity, the device resistive states increase, as required for reduced power consumption. An electric field‐driven TaOx oxidation/reduction is responsible for the ReRAM switching. To demonstrate the potential of the optimized TaOx/HfO2 devices, the training of a fully‐connected neural network on the Modified National Institute of Standards and Technology database dataset is simulated and benchmarked against a full precision digital implementation. A conductive TaOx material is developed to create a complementary metal‐oxide‐semiconductor‐compatible bilayer TaOx/HfO2 resistive random‐access memory. This device is based on filamentary switching mechanisms, but shows reduced stochasticity and improved graduality compared to metal–insulator–metal baselines. Applying short (<200 ns) and low amplitude (<1.5 V) voltage pulses the device responds with quasi‐analog conductance updates in both directions.
Author Stecconi, Tommaso
Bragaglia, Valeria
Horst, Folkert
Guido, Roberto
Popoff, Youri
Sousa, Marilyne
Drechsler, Ute
Berchialla, Luca
Offrein, Bert Jan
Dittmann, Regina
Dávila, Diana
Halter, Mattia
La Porta, Antonio
Weiss, Jonas
Author_xml – sequence: 1
  givenname: Tommaso
  orcidid: 0000-0001-5385-3486
  surname: Stecconi
  fullname: Stecconi, Tommaso
  email: tec@zurich.ibm.com
  organization: IBM Research GmbH‐Zurich Research Laboratory
– sequence: 2
  givenname: Roberto
  surname: Guido
  fullname: Guido, Roberto
  organization: IBM Research GmbH‐Zurich Research Laboratory
– sequence: 3
  givenname: Luca
  surname: Berchialla
  fullname: Berchialla, Luca
  organization: IBM Research GmbH‐Zurich Research Laboratory
– sequence: 4
  givenname: Antonio
  surname: La Porta
  fullname: La Porta, Antonio
  organization: IBM Research GmbH‐Zurich Research Laboratory
– sequence: 5
  givenname: Jonas
  surname: Weiss
  fullname: Weiss, Jonas
  organization: IBM Research GmbH‐Zurich Research Laboratory
– sequence: 6
  givenname: Youri
  surname: Popoff
  fullname: Popoff, Youri
  organization: ETH Zurich
– sequence: 7
  givenname: Mattia
  surname: Halter
  fullname: Halter, Mattia
  organization: ETH Zurich
– sequence: 8
  givenname: Marilyne
  surname: Sousa
  fullname: Sousa, Marilyne
  organization: IBM Research GmbH‐Zurich Research Laboratory
– sequence: 9
  givenname: Folkert
  surname: Horst
  fullname: Horst, Folkert
  organization: IBM Research GmbH‐Zurich Research Laboratory
– sequence: 10
  givenname: Diana
  surname: Dávila
  fullname: Dávila, Diana
  organization: IBM Research GmbH‐Zurich Research Laboratory
– sequence: 11
  givenname: Ute
  surname: Drechsler
  fullname: Drechsler, Ute
  organization: IBM Research GmbH‐Zurich Research Laboratory
– sequence: 12
  givenname: Regina
  surname: Dittmann
  fullname: Dittmann, Regina
  organization: RWTH Aachen University
– sequence: 13
  givenname: Bert Jan
  surname: Offrein
  fullname: Offrein, Bert Jan
  organization: IBM Research GmbH‐Zurich Research Laboratory
– sequence: 14
  givenname: Valeria
  surname: Bragaglia
  fullname: Bragaglia, Valeria
  organization: IBM Research GmbH‐Zurich Research Laboratory
BookMark eNpNkM9OwkAYxDdGExG5et4XKHz7r7THBkFIQBKCibfm67KLq-2WtEXk5iP4jD6JJRriaWaSySTzuyGXvvSGkDsGfQbAB2jyos-BcwApowvS4SyOAxbC8-U_f016df0KAGwYCqlEh2QTl2NhfIPVka5x-TGY2iWnK7NKFvTevDttamrLij6afYV5K82hrN5quq7Qeee39OCaF5p4zMstnfnvz6-FKcp2bFQWu33TNm7JlcW8Nr0_7ZKnyXg9mgbz5cNslMyDmksVBSrmCCbSGFrGQEVWqwwVixWG0oQbo7m2gJDJDYqhFjYWOkNptArbjlWZ6JL4d_fgcnNMd5Ur2lMpg_REKD0RSs-E0mQ8X5yT-AGf4WA1
ContentType Journal Article
Copyright 2022 The Authors. Advanced Electronic Materials published by Wiley‐VCH GmbH
Copyright_xml – notice: 2022 The Authors. Advanced Electronic Materials published by Wiley‐VCH GmbH
DBID 24P
DOI 10.1002/aelm.202200448
DatabaseName Wiley Online Library Open Access
DatabaseTitleList
Database_xml – sequence: 1
  dbid: 24P
  name: Wiley Online Library Open Access
  url: https://authorservices.wiley.com/open-science/open-access/browse-journals.html
  sourceTypes: Publisher
DeliveryMethod fulltext_linktorsrc
EISSN 2199-160X
EndPage n/a
ExternalDocumentID AELM202200448
Genre article
GrantInformation_xml – fundername: H2020 Industrial Leadership
  funderid: 871371; 871330
– fundername: H2020 Marie Skłodowska‐Curie Actions
  funderid: 861153
– fundername: Horizon 2020 Framework Programme
  funderid: 871391; 871658
GroupedDBID 0R~
1OC
24P
33P
AAESR
AAHHS
AAXRX
AAZKR
ABCUV
ACAHQ
ACCFJ
ACCMX
ACCZN
ACGFS
ACPOU
ACXBN
ACXQS
ADBBV
ADKYN
ADOZA
ADXAS
ADZMN
ADZOD
AEEZP
AENEX
AEQDE
AFBPY
AIACR
AIURR
AIWBW
AJBDE
ALMA_UNASSIGNED_HOLDINGS
ALUQN
AMYDB
ARCSS
AVUZU
AZVAB
BFHJK
BMXJE
BRXPI
DCZOG
EBS
EJD
GODZA
GROUPED_DOAJ
LATKE
LEEKS
LITHE
LOXES
LUTES
LYRES
MEWTI
O9-
P2W
ROL
SUPJJ
WBKPD
WOHZO
WXSBR
ZZTAW
ID FETCH-LOGICAL-s2458-592a0e8ca6f11058fc5ba5195a64e6dec2cf0a0b4da37c3f93cba4ec56195f5b3
IEDL.DBID 24P
ISICitedReferencesCount 23
ISICitedReferencesURI http://www.webofscience.com/api/gateway?GWVersion=2&SrcApp=Summon&SrcAuth=ProQuest&DestLinkType=CitingArticles&DestApp=WOS_CPL&KeyUT=000822534500001&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D
ISSN 2199-160X
IngestDate Wed Jan 22 16:23:20 EST 2025
IsDoiOpenAccess true
IsOpenAccess true
IsPeerReviewed true
IsScholarly true
Issue 10
Language English
License Attribution
LinkModel DirectLink
MergedId FETCHMERGED-LOGICAL-s2458-592a0e8ca6f11058fc5ba5195a64e6dec2cf0a0b4da37c3f93cba4ec56195f5b3
ORCID 0000-0001-5385-3486
OpenAccessLink https://onlinelibrary.wiley.com/doi/abs/10.1002%2Faelm.202200448
PageCount 14
ParticipantIDs wiley_primary_10_1002_aelm_202200448_AELM202200448
PublicationCentury 2000
PublicationDate October 2022
PublicationDateYYYYMMDD 2022-10-01
PublicationDate_xml – month: 10
  year: 2022
  text: October 2022
PublicationDecade 2020
PublicationTitle Advanced electronic materials
PublicationYear 2022
References 2017; 8
2013; 1
2019; 11
2018; 124
2020; 17
2016; 31
2020; 15
2019; 18
2020; 14
2020; 12
2012; 59
2015; 107
2016; 37
1998; 86
2017; 9
2018; 1
1971; 18
2017; 38
1995; 23
2017; 32
2020; 577
2014; 57
2011; 21
2012; 29
1992; 46
1958; 5
2019; 7
2018; 29
2004; 85
2009; 21
2019; 5
2011
2015; 521
2010
2016; 10
2006; 5
2017; 29
2003
2018; 65
2015; 8
2012; 107
2014; 115
2016; 6
2014; 105
2015; 25
2022
2015; 62
2021
2000; 77
2013; 31
2008; 48
2019
2016; 63
2011; 88
2014; 35
2016
2017; 19
2015
2014
2013
2008; 453
2016; 8
2014; 104
References_xml – year: 2011
– volume: 105
  year: 2014
  publication-title: Appl. Phys. Lett.
– volume: 1
  year: 2013
  publication-title: APL Mater.
– volume: 85
  year: 2004
– volume: 453
  start-page: 80
  year: 2008
  publication-title: Nature
– volume: 107
  start-page: 02
  year: 2015
  publication-title: Appl. Phys. Lett.
– volume: 577
  start-page: 641
  year: 2020
  publication-title: Nature
– year: 2021
– volume: 17
  start-page: 261
  year: 2020
  publication-title: Nat. Methods
– volume: 88
  start-page: 1129
  year: 2011
  publication-title: Microelectron. Eng.
– volume: 5
  start-page: 208
  year: 1958
  publication-title: J. Phys. Chem. Solids
– volume: 12
  year: 2020
  publication-title: ACS Appl. Mater.
– volume: 21
  start-page: 4487
  year: 2011
  publication-title: Adv. Funct. Mater.
– year: 2014
– volume: 59
  start-page: 2468
  year: 2012
  publication-title: IEEE Trans. Electron Devices
– volume: 29
  year: 2018
  publication-title: Nanotechnology
– volume: 1
  start-page: 22
  year: 2018
  publication-title: Nat. Electron.
– volume: 5
  start-page: 312
  year: 2006
  publication-title: Nat. Mater.
– volume: 8
  year: 2017
  publication-title: Nat. Commun.
– volume: 31
  year: 2016
  publication-title: Semicond. Sci. Technol.
– volume: 8
  start-page: 7191
  year: 2015
  publication-title: Materials
– volume: 5
  year: 2019
  publication-title: Adv. Electron. Mater.
– volume: 29
  start-page: 141
  year: 2012
  publication-title: IEEE Signal Process Mag.
– volume: 63
  start-page: 1524
  year: 2016
  publication-title: IEEE Trans. Electron Devices
– volume: 11
  year: 2019
  publication-title: Nanoscale
– volume: 37
  start-page: 1268
  year: 2016
  publication-title: IEEE Electron Device Lett.
– volume: 65
  start-page: 3229
  year: 2018
  publication-title: IEEE Trans. Electron Devices
– volume: 46
  start-page: 7157
  year: 1992
  publication-title: Phys. Rev. B
– volume: 23
  start-page: 20
  year: 1995
  publication-title: ACM SIGARCH Comput. Archit. News
– year: 2019
– start-page: 457
  year: 2016
– year: 2015
– volume: 62
  start-page: 1998
  year: 2015
  publication-title: IEEE Trans. Electron Devices
– start-page: 136
  year: 2010
  end-page: 137
– volume: 18
  start-page: 309
  year: 2019
  publication-title: Nat. Mater.
– year: 2022
  publication-title: Adv. Electron. Mater.
– start-page: 995
  year: 2013
  end-page: 1004
– volume: 29
  year: 2017
  publication-title: Adv. Mater.
– volume: 19
  start-page: 2
  year: 2017
  publication-title: Comput. Sci. Eng.
– volume: 10
  year: 2016
  publication-title: Front. Neurosci.
– start-page: 297
  year: 2015
  end-page: 300
– volume: 48
  start-page: 4729
  year: 2008
  publication-title: J. Appl. Phys.
– year: 2003
– volume: 25
  start-page: 3
  year: 2015
  publication-title: IEEE Trans. Appl. Supercond.
– volume: 21
  start-page: 2632
  year: 2009
  publication-title: Adv. Mater.
– start-page: 31.6.1
  year: 2011
  end-page: 31.6.4
– volume: 8
  year: 2016
  publication-title: Nanoscale
– year: 2016
– volume: 115
  year: 2014
  publication-title: J. Appl. Phys.
– volume: 31
  start-page: 35
  year: 2013
  end-page: 59
– volume: 104
  year: 2014
  publication-title: Appl. Phys. Lett.
– volume: 32
  year: 2017
  publication-title: Semicond. Sci. Technol.
– volume: 77
  start-page: 139
  year: 2000
  publication-title: Appl. Phys. Lett.
– volume: 18
  start-page: 507
  year: 1971
  publication-title: IEEE Trans. Circuit Theory
– volume: 1
  start-page: 333
  year: 2018
  publication-title: Nat. Electron.
– volume: 124
  year: 2018
  publication-title: J. Appl. Phys.
– volume: 9
  year: 2017
  publication-title: ACS Appl. Mater. Interfaces
– volume: 14
  start-page: 103
  year: 2020
  publication-title: Front. Neurosci.
– volume: 7
  year: 2019
  publication-title: APL Mater.
– volume: 15
  start-page: 529
  year: 2020
  publication-title: Nat. Nanotechnol.
– volume: 521
  start-page: 61
  year: 2015
  publication-title: Nature
– volume: 38
  start-page: 1019
  year: 2017
  publication-title: IEEE Electron Device Lett.
– volume: 57
  year: 2014
– volume: 37
  start-page: 994
  year: 2016
  publication-title: IEEE Electron Device Lett.
– volume: 107
  start-page: 509
  year: 2012
  publication-title: Appl. Phys. A: Mater. Sci. Process.
– volume: 6
  start-page: 146
  year: 2016
  publication-title: IEEE J. Emerging Sel. Top. Circuits Syst.
– volume: 86
  start-page: 2278
  year: 1998
  publication-title: Proc. IEEE
– volume: 35
  start-page: 912
  year: 2014
  publication-title: IEEE Electron Device Lett.
SSID ssj0001763453
Score 2.3976066
Snippet The in‐memory computing paradigm aims at overcoming the intrinsic inefficiencies of Von‐Neumann computers by reducing the data‐transport per arithmetic...
SourceID wiley
SourceType Publisher
SubjectTerms analog memory
artificial synapses
HfO 2
resistive random‐access memory
TaO x
Title Filamentary TaOx/HfO2 ReRAM Devices for Neural Networks Training with Analog In‐Memory Computing
URI https://onlinelibrary.wiley.com/doi/abs/10.1002%2Faelm.202200448
Volume 8
WOSCitedRecordID wos000822534500001&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
journalDatabaseRights – providerCode: PRVHPJ
  databaseName: ROAD: Directory of Open Access Scholarly Resources
  customDbUrl:
  eissn: 2199-160X
  dateEnd: 99991231
  omitProxy: false
  ssIdentifier: ssj0001763453
  issn: 2199-160X
  databaseCode: M~E
  dateStart: 20150101
  isFulltext: true
  titleUrlDefault: https://road.issn.org
  providerName: ISSN International Centre
link http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwpV1LS8NAEF60evDiAxXf7MFr6Gazm8exaEuFpi2lQm9hdjsrBYnSVLEX8Sf4G_0l7ia1tVe9hAQyIcxj52OY-YaQ6whCAwDCA2TgCcWMBzpinox0MMZY2ZxXDgp3om43Ho2S_q8p_oofYllwc5FRntcuwEEV9RVpKOCjmyTnzswi3iRbvh9Ezq-56K-qLDZ8RElFaSMz8fyQjX6YGxmvr39iHZ6W-aW19_8_2ye7C2xJG5UzHJANzA-Jak2s0V2H-HROh9B7q7dNj9MBDhopvcXyoKAWuVJH02Glu1VfeEGHi-UR1JVqqSMveXqgd_nXx2fqunPntFoIYd84Ivet5vCm7S0WK3gFFzL2ZMKBYaytnWz2l7HRUoGjmYFQYDhGzbVhwJQYQ2BtZpJAKxCoLdZKpJEqOCa1_CnHE0IjI60bgI8cIoFGxsqBOo2-Qiun2Snhpaqy54o8I6toknnmlJQtlZQ1mp10-XT2F6FzsuPuq0a7C1KbTV_wkmzr19mkmF6VbmGv6XvzGw94vJs
linkProvider Wiley-Blackwell
linkToHtml http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwrV1LSwMxEA5aBb34QMW3OXgNTdNkH8eiLRV3t6Ws0NsySRMpyFbaKvbmT_A3-ktMdteWXsXjLjthmUdmGL75BqFbHzwDAJyApkC4pIaA8ikRvmqOdCBtzisGhSM_SYLhMOxXaEI3C1PyQywbbi4yivvaBbhrSNdXrKGgX9woOXN25sEm2uI21bglBoz3V20WGz-84KK0oRmShkeHv9SNlNXXj1ivT4sE09n_h187QHtVdYlbpTscog2dHyHZGVuzO4z4dIFT6H3Uu6bH8EAPWjG-18VVgW3tih1Rh5VOSmT4DKfV-gjsmrXY0ZdMnvFD_v35FTt87gKXKyHsF8foqdNO77qkWq1AZoyLgIiQAdWBspay-V8ERgkJjmgGPK69kVZMGQpU8hE0rdVM2FQSuFa22gqFEbJ5gmr5JNenCPtGWEeAhmbgc21EIF1Zp3RDaiun6Bliha6y15I-IyuJklnmlJQtlZS12lG8fDr_i9AN2ummcZRFD8njBdp170vY3SWqzadv-gptq_f5eDa9LnzkB9krv0w
linkToPdf http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwrV1LSwMxEA5aRbz4QMW3OXgNTdNkH8diWyr2RanQ2zLJTqQgbWmr2Js_wd_oLzHZrS29iscNOyHMm2HmG0LuQwgsAEgGyIFJzS0DE3KmQlNOMdIu5mWDws2w3Y4Gg7i77Cb0szA5PsSq4OYtI_PX3sBxktriGjUU8NWPkgsvZxltkx2pnKP14M6yuy6zOPuRGRalM82YlQI--IVu5KK4ecVmfpoFmPrhPzztiBwss0taydXhmGzh6ITo-tCJ3feITxe0D52PYsN2BO1hr9KiVcxcBXW5K_VAHY66nXeGz2h_uT6C-mIt9fAl4xf6OPr-_Gr5_twFzVdCuD9OyXO91n9osOVqBTYTUkVMxQI4RsZJysV_FVmjNHigGQgkBikaYSwHrmUKZSc1G5eNBonGZVuxskqXz0hhNB7hOaGhVU4RoIQCQolWRdqndQZLGh2d4RdEZLxKJjl8RpIDJYvEMylZMSmp1Jqt1dflX4juyF63Wk-aj-2nK7Lvj_Ouu2tSmE_f8Ibsmvf5cDa9zVTkB9X2vtA
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=Filamentary+TaOx%2FHfO2+ReRAM+Devices+for+Neural+Networks+Training+with+Analog+In%E2%80%90Memory+Computing&rft.jtitle=Advanced+electronic+materials&rft.au=Stecconi%2C+Tommaso&rft.au=Guido%2C+Roberto&rft.au=Berchialla%2C+Luca&rft.au=La+Porta%2C+Antonio&rft.date=2022-10-01&rft.issn=2199-160X&rft.eissn=2199-160X&rft.volume=8&rft.issue=10&rft.epage=n%2Fa&rft_id=info:doi/10.1002%2Faelm.202200448&rft.externalDBID=10.1002%252Faelm.202200448&rft.externalDocID=AELM202200448
thumbnail_l http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=2199-160X&client=summon
thumbnail_m http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=2199-160X&client=summon
thumbnail_s http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=2199-160X&client=summon